CN102243302A - Method for extracting line spectrum time accumulation characteristics of hydro-acoustic target radiation noise - Google Patents

Abstract

The invention discloses a method for extracting line spectrum time accumulation characteristics of a hydro-acoustic target radiation noise. The method comprises the following steps of: performing characteristic extraction by using a low-frequency array (LOFAR) diagram of the line spectrum time accumulation characteristics; and by time sequence analysis of a sonar signal and information and using processing methods of stable line spectrum determination, correlation detection, line spectrum drifting detection and the like, digging and extracting characteristic information which is correlated with a target behavior and the like. Compared with a single extraction method of line spectrum characteristics, the method provided by the invention has the advantages that: by using the line spectrum time accumulation characteristics, some characteristics of a target can be accumulated and continued in terms of time to reflect the change situation of the characteristics in terms of time, so the steady-state characteristics of the target can be more stable and more reliable, the dynamic characteristics and transient-state characteristics of the target can be extracted, and reference is provided for the classification and identification of the target.

Description

A kind of line spectrum accumulated time feature extracting method of underwater acoustic target radiated noise

Technical field

The present invention relates to the automatic Classification and Identification technology of Underwater Acoustic Object, relate in particular to a kind of line spectrum accumulated time feature extracting method of underwater acoustic target radiated noise.

Background technology

The target classification identification of sonar mainly is that surface vessel, underwater submarine, torpedo and other immersed bodies are carried out Classification and Identification.Because various naval vessels, different models, different operating modes and under different ocean conditions shown characteristic of coming out all different, make target property be difficult to describe and Classification and Identification.So the automatic Classification and Identification of Underwater Acoustic Object is crucial problem in the Underwater Acoustics Engineering, also be a Underwater acoustic signal processing difficult problem of generally acknowledging.

Sonar is divided into passive sonar and active sonar from system, also is divided into passive sonar identification and active sonar and discerns two classes in target classification identification.Passive sonar identification as methods such as Fourier transform, modulation, wavelet transformations, is extracted feature by the underwater acoustic target radiated noise that receives is carried out analyzing and processing the signal after conversion, realize the Classification and Identification of target.But because the influence that neighbourhood noise, interference and battle array are handled, and the complicacy of target radiated noise such as naval vessel etc., make many methods to the Classification and Identification of the actual sonar received signal in sea the time, be difficult to obtain desirable effect.In the past, the research of line spectrum often was confined to real-time, and had ignored its relevance in time.The line spectrum that single extracts is very unstable, and this shows two aspects: the one, and true line spectrum is by noise takeover, thereby can't extract from power spectrum, so just produced and failed to judge; The 2nd, the very noisy line spectrum appears, and it is mistaken for the line spectrum of target, has caused void to be declared.

So-called line spectrum accumulated time feature is accumulated the time-frequency two-dimensional feature of formation exactly in time to line spectrum.Utilize long information accumulation can improve the degree of stability of information undoubtedly, the LOFAR spectrogram shows or analysis so all designed in many passive sonars.But mostly still as a kind of display message, LOFAR spectrogram target's feature-extraction method is less automatically.

Line spectrum accumulated time feature is as a kind of temporal aspect, its objective is and some feature of target is accumulated in time and continue, reflect these features situation of change in time whereby, thereby can make the steady state characteristic of target stable more and reliable, also can therefrom extract the behavioral characteristics and the transient state characteristic of target, for the Classification and Identification of target provides foundation.

Summary of the invention

The objective of the invention is to overcome the shortcoming that single extracts the line spectrum characterization method, a kind of underwater acoustic target radiated noise line spectrum accumulated time Feature Extraction method is provided, utilize line spectrum time accumulative total feature, promptly utilizing the LOFAR spectrogram to carry out target signature extracts automatically, by time series analysis to sonar signal and information, obtain distribution and the variation of line spectrum on frequency, therefrom extract clarification of objective information, thereby help identification.

For achieving the above object, the present invention takes following technical scheme: a kind of line spectrum accumulated time feature extracting method of underwater acoustic target radiated noise, it is characterized in that: the underwater acoustic target radiated noise signal is carried out short time discrete Fourier transform, obtain line spectrum accumulated time characteristic pattern, by time series analysis to sonar signal and information, extract target line spectrum number and line spectrum drift features, comprise following process:

A. obtain line spectrum accumulated time characteristic pattern

A1 establishes and gathers the underwater acoustic target radiated noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sample frequency of radiated noise signals sequence is f _s, N 〉=f _s, data are carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n);

A2 is by bandpass filter, to S ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n);

A3 is to s ₂(n) carry out short time discrete Fourier transform, obtain line spectrum accumulated time feature LOFAR figure S (i, j), i=1 wherein, 2 ..., M, M are that time discrete is counted, j=1, and 2 ..., to be that frequency domain is discrete count for N, N;

B. extract and stablize the line spectrum number

The method that adopts statistics from line spectrum accumulated time feature LOFAR figure S (i extracts the line spectrum number in j), eliminates the interference of random noise, obtains reliable and stable line spectrum number:

B1. (i j) carries out accumulated time, obtains the histogram line spectrum to S

$Z=\left\{z_j\right\}=\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}S(i,j),j=\mathrm{0,1,2},...,N$

Z wherein _jBe the histogram line spectrum amplitude of j frequency, Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

B2. set line spectrum decision threshold gate, the value of gate is three times of histogram amplitude equalizing value;

B3. for line spectrum j point, if z _j〉=gate, then the line spectrum number adds 1, h _j=1; Otherwise, if z _j＜gate, then h _j=0;

B4. obtaining stablizing line spectrum distributes J=1,2 ..., to be that frequency domain is discrete count for N, N;

C. correlation detection

Adopt correlation detection, the non-target line spectrum that exists during the stable line spectrum that obtains is distributed is got rid of, to obtain target line spectrum number:

C1. calculate original power spectral sequence E=S ²(i, j)={ w _Ij, i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

C2. k bar line spectrum at the power of moment i is in the stable line spectrum that has got

S wherein, l are power summation scope, then the power sequence W of k bar line spectrum _kFor

W _k＝[w _k(1)，…，w _k(Wide)]，k＝1，2，…，N

Wherein Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

C3. the power sequence correlativity of calculating line spectral power sequence between in twos:

Average $m_k=\frac{1}{\mathrm{Wide}}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{w}_k\left(i\right)$

Variance ${{\mathrm{\σ}}_k}^2=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{(w_k\left(i\right)-m_k)}^2$

Covariance $C_{k_1{k}_2}=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}(w_{k_1}\left(i\right)-m_{k_1})(w_{k_2}\left(i\right)-m_{k_2})$

Related coefficient ${\mathrm{\ρ}}_{k_1{k}_2}=\frac{C_{k_1{k}_2}}{{\mathrm{\σ}}_{k_1}{\mathrm{\σ}}_{k_2}}$

Wherein Wide is that time discrete is counted, if the related coefficient of two line spectrum power sequences greater than 0.5, is then thought line spectrum of the same type;

C4. judge whether to be the target line spectrum, thresholding gate1 is set, and gate1 is 0.6, for when the variance of power sequence greater than gate1, just think the target line spectrum;

C5. the assignment of target line spectrum is 1 with score spectral position assignment, and all the other positions are 0;

Target line spectrum spectrum distribution

J=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D. the line spectrum drift detects

By analyzing the drift rule of line spectrum, draw the motion feature information of target:

D1. calculating i target line spectrum constantly distributes:

I target line spectrum distribution H constantly _i=S (i, j) * xp ', i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D2. calculating the two dimension target line spectrum distributes:

XP=[H ₁, H ₂..., H _Wide]={ h _Ij,

I=1,2 ..., Wide, j=0,1,2 ..., N wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D3. denoising:

Work as h _Ij=1, during 2≤i≤Wide-1, then judge

With

Whether be 0, if be 0, h then _Ij=0, wherein w is the hunting zone, and Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

D4. seek the line spectrum central point:

For the value of i target line spectrum distribution Hi constantly, if continuous 1, then the line spectrum sign is integrated into central point, and assignment is 1, all the other some assignment are 0;

D5. the search of drifting about:

H from the moment 1 _1j=1, begin search, successively follow-up the 2nd, 3 ..., on the frequency spectrum of Wide period, search in spectrum width [w, the w] scope that limits supposes that the position of the line spectrum existence of last time is j, if searched h in [w, w] _{I (j-k)}=1, k≤w notes position j '=j-k, and adds up the number of times that searches; Otherwise j '=j continues search on next frequency spectrum constantly.Until searching last constantly.

By above-mentioned 4 concrete steps, the line spectrum accumulated time characteristic information of extraction provides foundation for the Classification and Identification of target.

Compared with prior art, the present invention has following advantage:

1) the present invention proposes and utilize line spectrum accumulated time characteristic pattern to carry out feature extraction, line spectrum accumulated time feature is as a kind of temporal aspect, its objective is and some feature of target is accumulated in time and continue, reflect these features situation of change in time whereby, thereby can make the steady state characteristic of target stable more and reliable, also can therefrom extract the behavioral characteristics and the transient state characteristic of target, for the Classification and Identification of target provides foundation.The present invention accumulates in time to the line spectrum that single extracts, a kind of time-frequency two-dimensional feature that extracts on this basis.Compared to the line spectrum that single extracts, line spectrum accumulated time feature has that the noise of reduction is strong, property time correlation is strong, contain superiority such as target dynamic information.

2) the invention provides the decision method of target line spectrum number, at first stablize the judgement of line spectrum number, and utilize correlation detection, calculate the correlativity of power spectrum between different line spectrums, classify stablizing line spectrum.And then according to the characteristic of dissimilar sound-source signals, distinguish the target line spectrum and disturb line spectrum, judge target line spectrum number.

3) the invention provides target line spectrum drift detection method, different target is owing to the difference of motion feature produces different Doppler frequency deviation effects, thereby provides foundation for Classification and Identification.Whether doppler shift effect shows as the drift of frequency on power spectrum, exist continuous drift further to judge Doppler shift by detecting the target line spectrum.

Description of drawings

Fig. 1 is a line spectrum accumulated time feature extracting method block diagram of the present invention;

Fig. 2 is that line spectrum accumulated time characteristic pattern (LOFAR figure) obtains process flow diagram;

Fig. 3 is the LOFAR figure of the line spectrum accumulated time feature of certain target, and horizontal ordinate is a frequency, and ordinate is the time;

Fig. 4 is the histogram of the line spectrum accumulated time feature of certain target;

Fig. 5 is the Doppler shift figure of the different initial angles of identical speed;

Fig. 6 is the Doppler shift figure at fundamental frequency of different target;

Fig. 7 is the serpentine locomotion trajectory diagram;

Doppler shift when Fig. 8 is serpentine locomotion.

Below in conjunction with accompanying drawing the present invention is done further explanation.

Referring to Fig. 1-3, a kind of line spectrum accumulated time feature extracting method of underwater acoustic target radiated noise has been described, method comprises the steps such as judgement, correlation detection and line spectrum drift detection that line spectrum time accumulative total characteristic pattern (LOFAR figure) calculated, stablized the line spectrum number:

A. obtain line spectrum accumulated time characteristic pattern

A1 establishes and gathers the underwater acoustic target radiated noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sample frequency of radiated noise signals sequence is f _s, N 〉=f _s, data are carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n);

A2 is by bandpass filter, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n);

A3 is to s ₂(n) carry out short time discrete Fourier transform, obtain line spectrum accumulated time feature LOFAR figure S (i, j), i=1 wherein, 2 ..., M, M are that time discrete is counted, j=1, and 2 ..., to be that frequency domain is discrete count for N, N;

B. extract and stablize the line spectrum number

Because different targets, the difference of its power system, Ship Structure and self character makes that radiated noise separately is also just different on line spectrum and continuous spectrum structure.Through a large amount of statistical observations, find in the power spectrum of many different underwater acoustic target signals that there is certain difference the position that line spectrum quantity and line spectrum occur.Adopt the method for statistics from LOFAR figure, to extract the line spectrum number, just can eliminate the interference of random noise, obtain reliable and stable line spectrum number:

B1. (i j) carries out accumulated time, obtains the histogram line spectrum to S

$Z=\left\{z_j\right\}=\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}S(i,j),$ j＝0，1，2，…，N

Z wherein _jBe the histogram line spectrum amplitude of j frequency, Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

B2. set line spectrum decision threshold gate, the value of gate is three times of histogram amplitude equalizing value;

B3. for line spectrum j point, if z _j〉=gate, then the line spectrum number adds 1, h _j=1; Otherwise, if z _j＜gate, then h _j=0;

B4. obtain stablizing line spectrum distribution xp={h _j,

J=1,2 ..., to be that frequency domain is discrete count for N, N;

C. correlation detection

Though can obtain stablize the line spectrum number by front one step, but in these stable line spectrums, still exist non-target line spectrum, as the interference of this warship and environment steady noise etc.The existence of this non-target line spectrum has influenced reliability and confidence level for Target Recognition.The present invention removes non-target line spectrum by adopting correlation detection:

C1. because line spectrum accumulated time feature can write down the power sequence of the line spectrum that extracts in the certain hour.So,,, calculate original power spectral sequence E=S according to its place frequency for the line spectrum that extracts ²(i, j)={ w _Ij, i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

C2. k bar line spectrum at the power of moment i is in the stable line spectrum that has got

S wherein, l are power summation scope, then the power sequence W of k bar line spectrum _kFor

Wk＝[w _k(1)，…，w _k(Wide)]，k＝1，2，…，N

Wherein Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

C3. the correlativity of power is different between the different line spectrum types, the correlativity height of power between the same type line spectrum, and the correlativity of power is just low between dissimilar line spectrums.Therefore need the power sequence correlativity of calculating line spectral power sequence between in twos:

Average $m_k=\frac{1}{\mathrm{Wide}}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{w}_k\left(i\right)$

Variance ${{\mathrm{\σ}}_k}^2=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{(w_k\left(i\right)-m_k)}^2$

Covariance $C_{k_1{k}_2}=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}(w_{k_1}\left(i\right)-m_{k_1})(w_{k_2}\left(i\right)-m_{k_2})$

Related coefficient ${\mathrm{\ρ}}_{k_1{k}_2}=\frac{C_{k_1{k}_2}}{{\mathrm{\σ}}_{k_1}{\mathrm{\σ}}_{k_2}}$

Wherein Wide is that time discrete is counted.After the correlativity of asking between the line spectrum power sequence, by comparison, just line spectrum can be divided into several classes, again if the related coefficient of two line spectrum power sequences greater than 0.5, is then thought line spectrum of the same type to power related coefficient between all line spectrums;

C4. owing to dissimilar stable line spectrums, its power situation of change in time is different.This warship disturbs for receiver, and its distance is constant relatively, so its power is relatively stable in time; And even relative velocity is stable, there is not the line spectrum offset that causes because of Doppler effect because relative receiver moves in target, and the power of target line spectrum also changes; The power of integrated environment steady noise then is stochastic distribution.Therefore judge when whether line spectrum is the target line spectrum that thresholding gate1 can be set, and gate1 is 0.6, when the variance of power sequence greater than gate1, just think the target line spectrum;

C5. the assignment of target line spectrum is 1 with score spectral position assignment, and all the other positions are 0;

Target line spectrum spectrum distribution xp '={ h _j,

J=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N.

D. the line spectrum drift detects

We can see from histogram, all have certain expansion near the centre frequency of line spectrum, just show as the drift of line spectrum on LOFAR figure, and line spectrum drift packets of information contains the motion feature of target.By analyzing the drift rule of line spectrum, can draw the motion feature information of target, help to improve the accuracy rate of Target Recognition.The concrete steps that the line spectrum drift detects are as follows:

D1. calculating i target line spectrum constantly distributes:

I target line spectrum distribution H constantly _i=S (i, j) * xp ', i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D2. calculating the two dimension target line spectrum distributes:

XP=[H ₁, H ₂..., H _Wide]={ h _Ij,

I=1,2 ..., Wide, j=0,1,2 ..., N wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D3. because may there be the noise point in two-dimentional line spectrum in distributing, the target line spectrum distributes generally certain width, if but have isolated peak point, then this point is exactly a noise, need carry out denoising:

D4. seek the line spectrum central point:

For i target line spectrum distribution H constantly _iValue, if continuous 1, then the line spectrum sign is integrated into central point, and assignment is 1, all the other some assignment are 0;

D5. because Doppler shift shows as the drift of frequency on power spectrum, can whether exist continuous position change to judge that Doppler shift changes by line spectrum.Therefore from 1 h constantly _1j=1, begin search, successively follow-up the 2nd, 3 ..., on the frequency spectrum of Wide period, search in spectrum width [w, the w] scope that limits supposes that the position of the line spectrum existence of last time is j, if searched h in [w, w] _{I (j-k)}=1, k≤w notes position j '=j-k, and adds up the number of times that searches; Otherwise j '=j continues search on next frequency spectrum constantly, until searching last constantly.The line spectrum drift features that search obtains can be carried out the judgement of target travel situation for the back, as have stable Doppler shift, and also promptly have under the stable line spectrum figure situation, be the target of linear uniform motion, and direction of motion and receiver are promptly constant in radial velocity on same straight line; Doppler shift changes little, is low speed or the less target of course angle θ variation; Doppler shift changes under the comparatively tangible situation, is that high speed or course angle θ change target greatly; Frequently wafing about continuous can appear in line spectrum, is hunting, because serpentine locomotion is the peculiar motor pattern of torpedo target, in case detect this special frequency change, just can identify this type of target.

Fig. 4 is the histogram that is depicted as in the line spectrum accumulated time feature at the line spectrum accumulated time feature of certain target, as can be seen from the figure, sets thresholding, can rule out stable line spectrum number, and can determine the position of line spectrum.

Fig. 5 is the Doppler shift of the different initial angles of identical speed.What show among the figure is at initial distance R ₀=10000m, initial angle is respectively θ ₁=30 ° (dotted line) and θ ₂Because changing, radial velocity causes the situation of frequency change when=60 ° (solid line) constantly in difference.Flex point among the figure is the situation of target travel direction when vertical with receiver.Along with the propelling of time, can see that frequency is in continuous skew.From in essence, the difference of initial angle is not remarkable to the influence of doppler shift effect.Because what influence Doppler shift here is radial velocity, and with not tangible contact of angle angle of distance distance or target travel direction and receiver line.From figure, also as can be seen, little initial angle can be regarded as a kind of delay in time of big initial angle.

Fig. 6 is that different target is with the Doppler shift of friction speed under identical initial angle.In this case, different target has different doppler shift effects, and just the Classification and Identification for target provides a kind of possibility.0000 demonstration is at initial distance R among the figure ₀=10000m, during initial angle θ=30 °, the fundamental frequency doppler shift effect of high-speed target (dotted line) and slower-velocity target (solid line).Flex point among the figure also is the situation of target travel direction when vertical with receiver.As can be seen from the figure, high-speed target has than the more obvious frequency shift effect of slower-velocity target.

Fig. 7 is the serpentine locomotion track.For some special objective, has special forms of motion because of special purposes.As serpentine curve motion etc., its equation of motion is more complicated, and doppler shift effect also has bigger difference with linear uniform motion.For serpentine locomotion, though the direction of motion of target is in continuous change, but can think that the whole course of target is constant substantially, mass motion speed also can be thought constant, thereby serpentine locomotion can be regarded as in the target uniform motion process, and the stack disturbance causes.To simplify the analysis, can think that this disturbance is frequency and all constant sinusoidal perturbation of amplitude.Therefore, serpentine locomotion can be regarded as sinusoidal perturbation vertical with its direction of motion of linear uniform motion stack.Track when empty straight line is for no disturbance among the figure, curve is the serpentine locomotion track after the sinusoidal perturbation that superposeed, target is driven towards the initial point direction by the distant place.

As can be seen from Figure 8, the Doppler shift of special objective has the obvious variation rate, with rectilinear motion significant difference is arranged also.Because serpentine locomotion is the peculiar motor pattern of torpedo target, so, in case detect this special frequency change, just can identify this type of target.Utilize the doppler shift effect of target can be to target Classification and Identification in addition, key in application be how to extract Doppler shift information from information source.

The present invention is directed to the Underwater Acoustic Object automatic recognition problem, propose to utilize line spectrum accumulated time feature LOFAR figure to carry out feature extraction, line spectrum accumulated time feature is carried out line spectrum count judgement, correlation detection, line spectrum drift detection.The method that the present invention adopts is improved the feature confidence level of extraction, has also excavated some new target type features simultaneously.

Claims (1)

1. the line spectrum accumulated time feature extracting method of a underwater acoustic target radiated noise, it is characterized in that: the underwater acoustic target radiated noise signal is carried out short time discrete Fourier transform, obtain line spectrum accumulated time characteristic pattern, by time series analysis to sonar signal and information, extract target line spectrum number and line spectrum drift features, comprise following process:

A. obtain line spectrum accumulated time characteristic pattern

A1 establishes and gathers the underwater acoustic target radiated noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sample frequency of radiated noise signals sequence is f _s, N 〉=f _s, data are carried out standardization, E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n);

A2 is by bandpass filter, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n);

A3 is to s ₂(n) carry out short time discrete Fourier transform, obtain line spectrum accumulated time feature LOFAR figure S (i, j), i=1 wherein, 2 ..., M, M are that time discrete is counted, j=1, and 2 ..., to be that frequency domain is discrete count for N, N;

B. extract and stablize the line spectrum number

The method that adopts statistics from line spectrum accumulated time feature LOFAR figure S (i extracts the line spectrum number in j), eliminates the interference of random noise, obtains reliable and stable line spectrum number:

B1. (i j) carries out accumulated time, obtains the histogram line spectrum to S

$Z=\left\{z_j\right\}=\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}S(i,j),$ j＝0，1，2，…，N

Z wherein _jBe the histogram line spectrum amplitude of j frequency, Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

B2. set line spectrum decision threshold gate, the value of gate is three times of histogram amplitude equalizing value;

B3. for line spectrum j point, if z _j〉=gate, then the line spectrum number adds 1, h _j=1; Otherwise, if z _j＜gate, then h _j=0;

B4. obtain stablizing line spectrum distribution xp={h _j, J=1,2 ..., to be that frequency domain is discrete count for N, N;

C. correlation detection

Adopt correlation detection, the non-target line spectrum that exists during the stable line spectrum that obtains is distributed is got rid of, to obtain target line spectrum number:

C1. calculate original power spectral sequence E=S ²(i, j)={ w _Ij, i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

C2. k bar line spectrum at the power of moment i is in the stable line spectrum that has got

S wherein, l are power summation scope, then the power sequence W of k bar line spectrum _kFor

W _k＝[w _k(1)，…，w _k(Wide)]，k＝1，2，…，N

Wherein Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

C3. the power sequence correlativity of calculating line spectral power sequence between in twos:

Average $m_k=\frac{1}{\mathrm{Wide}}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{w}_k\left(i\right)$

Variance ${{\mathrm{\σ}}_k}^2=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}{(w_k\left(i\right)-m_k)}^2$

Covariance $C_{k_1{k}_2}=\frac{1}{\mathrm{Wide}-1}\underset{i=1}{\overset{\mathrm{Wide}}{\mathrm{\Σ}}}(w_{k_1}\left(i\right)-m_{k_1})(w_{k_2}\left(i\right)-m_{k_2})$

Related coefficient ${\mathrm{\ρ}}_{k_1{k}_2}=\frac{C_{k_1{k}_2}}{{\mathrm{\σ}}_{k_1}{\mathrm{\σ}}_{k_2}}$

Wherein Wide is that time discrete is counted, if the related coefficient of two line spectrum power sequences greater than 0.5, is then thought line spectrum of the same type;

C4. judge whether to be the target line spectrum.Thresholding gate1 is set, and gate1 is 0.6, for when the variance of power sequence greater than gate1, just think the target line spectrum;

C5. the assignment of target line spectrum is 1 with score spectral position assignment, and all the other positions are 0;

Target line spectrum spectrum distribution

J=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D. the line spectrum drift detects

By analyzing the drift rule of line spectrum, draw the motion feature information of target:

D1. calculating i target line spectrum constantly distributes:

I target line spectrum distribution H constantly _i=S (i, j) * xp ', i=1,2 ..., Wide, j=1,2 ..., N, wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D2. calculating the two dimension target line spectrum distributes:

XP=[H ₁, H ₂..., H _Wide]={ h _Ij,

I=1,2 ..., Wide, j=0,1,2 ..., N wherein Wide is that time discrete is counted, to be that frequency domain is discrete count N;

D3. denoising:

Work as h _Ij=1, during 2≤i≤Wide-1, then judge

With

Whether be 0, if be 0, h then _Ij=0, wherein w is the hunting zone, and Wide is that time discrete is counted, and to be that frequency domain is discrete count N;

D 4. seeks the line spectrum central point:

For i target line spectrum distribution H constantly _iValue, if continuous 1, then the line spectrum sign is integrated into central point, and assignment is 1, all the other some assignment are 0;

D 5. search of drifting about

H from the moment 1 _1j=1, begin search.Successively follow-up the 2nd, 3 ..., on the frequency spectrum of Wide period, search in spectrum width [w, the w] scope that limits supposes that the position of the line spectrum existence of last time is j, if searched h in [w, w] _{I (j-k)}=1, k≤w notes position j '=j-k, and adds up the number of times that searches; Otherwise j '=j continues search on next frequency spectrum constantly, until searching last constantly.

Images